The human visual system can receive multiple visual stimuli simultaneously from the environment, but only a limited amount of information can be further processed into the working memory as memory representations, in order to guide the follow-up cognitive activities. Previous fMRI studies have found that multiple brain regions including the front-parietal network and the temporal occipital region, all involved in the maintenance of visual working memory representations. And how do these brain regions maintain the memory representations during the gradual increase in memory load (i.e., number of memory items)?
The first study used scalp EEG, taking the advantage of its higher time resolution,explored how different brain regions maintain the representations of multiple working memory in time domain, and found that the frontal lobe, parietal lobe, temporal lobe and occipital lobe were all involved in the maintenance. The effect of memory load on the EEG amplitude of these brain regions changed dynamically. Only EEG activity in the late delay stage could better predict individual's performance. In the second study,intracranial EEG technology with higher temporal and spatial resolution was further adopted, which added rich frequency information. We found that the frontal lobe showed continuous enhancement of high-frequency neural activity in the maintenance stage of working memory representations, and the parietal lobe had sustained negative activation at lower frequencies. At lower frequencies, the sensory region first exhibited a significant negative activation, then switched to a significant positive activation, which was regulated by working memory load and embodied the function role of low-frequency power, that was gating by inhibition. The third study explored the functional connectivity between different brain regions during representations' maintenance. The results showed a significant delta band phase synchronization, which was also modulated by the working memory load, with higher load level induced longer phase synchronization. Finally in the fourth study, we did a cross frequency coupling analysis and found a significant load effect on the delta band phase and gamma band amplitude coupling in parietal, with lower load level inducing stronger coupling. This result was in line with the nested neuronal oscillation model. Therefore, this model provided a good interpretation of the neural activity during working memory representations' maintenance.
In general, multiple brain regions participated in the complex cognitive process of multiple memory representations, constructed a distributed brain network with strong dynamics in both the time and frequency domain, and communicated with each other through low-frequency phase synchronization. The phase amplitude coupling of the parietal activity was in line with the nested neuronal oscillation model